[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]


> From: Anant Kabra valero com [mailto:Anant Kabra valero com]
> Just a quick question....
> How does this compare to the NGPT effort by IBM??

The major highlights are basically these: NGPT uses a non-preemptive M:N
model, that implies a user level scheduler, a thicker wrapper over some
things [specially blocking system calls], where NPLT uses 1:1, preemptive,
where most of the ugly work is done by the kernel and the wrapper is really

The main discussion point on M:N vs 1:1 is that thread switch is faster in
M:N [switch to other thread's context] than on 1:1 [switch to kernel, select
next, switch to thread] (add to this some other things on the user context
switch being lighter than the kernel context switch). However the thicker
wrapper implies overheads that also need to be considered depending on the
application [ie: CPU bound stuff vs. IO bound vs. mixed workloads, etc ...]

Interface wise, they should be the same [at least on the long term when
finished], internally, it is a whole different story.

Inaky Perez-Gonzalez -- I do not speak for Intel Corp - opinions are solely
my own [or my fault]

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]